Cathodoluminescence Nano-Characterization of Semiconductors
Total Page:16
File Type:pdf, Size:1020Kb
Cathodoluminescence nano-characterization of semiconductors Paul R. Edwards and Robert W. Martin Department of Physics, SUPA, University of Strathclyde, 107 Rottenrow, Glasgow G4 0NG, UK Email: [email protected], [email protected] Abstract. We give an overview of the use of cathodoluminescence (CL) in the scanning electron microscope (SEM) for the nano-scale characterization of semiconducting materials and devices. We discuss the technical aspects of the measurement, such as factors limiting the spatial resolution and design considerations for efficient collection optics. The advantages of more recent developments in the technique are outlined, including the use of the hyperspectral imaging mode and the combination of CL and other SEM-based measurements. We illustrate these points with examples from our own experience of designing and constructing CL systems and applying the technique to the characterization of III-nitride materials and nanostructures. PACS: 78.60.Hk, 68.37.Hk 1. Introduction Cathodoluminescence (CL) is the phenomenon of light emission from a material under excitation by an energetic electron beam. The use of CL as an image-forming mode in scanning electron microscopy (SEM) can be traced back to work carried out on the first, pre- commercial, SEMs [1]. While early applications were mainly restricted to the study of cathode ray tube phosphors and geological samples, the technique began to be used for the analysis of epitaxial semiconducting materials during the 1960s [2]. The development of the technique as a characterization tool was driven by a number of differences between CL and the analogous techniques of photo- and electroluminescence (PL, EL). The principal advantage of CL lies in the fact that the spatial resolution is determined by the distribution of excess carriers in the material, and is not therefore limited by diffraction in the collection and excitation optics inherent in all far-field techniques. This has allowed the technique to be pushed beyond the micron scale, and this paper will discuss its application to the nanometre- scale characterization of semiconductor materials and devices. The generation of electron-hole pairs in a semiconductor under irradiation by energetic electrons occurs through the mechanism of impact ionization. Since the electron beam energy Ebeam used in the SEM will be typically between 3 and 4 orders of magnitude higher than the bandgap energy Eg of the semiconductor, this will result in a carrier generation rate far in excess of the incident electron flux. Also, while not allowing selective excitation of transitions below a given energy, as in PL excitation spectroscopy, this feature does allow all transitions to be excited even for the widest bandgap materials such as diamond and AlN. One further dissimilarity between CL, PL and EL is the potential for very different carrier injection levels. The small volume in which carriers are generated in CL can result in the carrier density exceeding the equilibrium levels in the semiconductor even at relatively low Cathodoluminescence nano-characterization of semiconductors beam currents. This provides a useful tool in CL characterization, with the current- dependence of the emission spectrum providing information on effects such as band-filling, for example. However, it also presents a possible pitfall when the effect is not taken into consideration, and care is not taken to maintain low injection conditions. The spectral, spatial, temporal, angular and polarization-dependent properties of cathodoluminescence are all routinely resolved, with the results being interpretable along broadly similar lines to those from other luminescence techniques. Emission lines associated with bound excitonic states, donor-acceptor pair bands and defect-related features typically dominate the spectra, with variations due to changes in (amongst other things) alloying, strain, doping and carrier concentration all providing insights into the material properties. CL images show the spatial distribution of these phenomena, while also allowing the distribution of defects to be visualized due to their higher rates of non-radiative recombination, and allowing micro- and nano-scale optoelectronic devices to be directly imaged. 2. Technical aspects 2.1. Spatial resolution The maximum spatial resolution achievable with CL is of particular interest, as it is this quantity which gives CL its prime advantage over alternative luminescence techniques. Factors potentially influencing the resolution include: the width of electron beam when it hits the sample surface; the volume of sample within which electron hole pairs are generated; and the subsequent movement of these charge carriers prior to their radiative recombination. 2.1.1 Beam size. While the minimum spot size achievable in modern field emission gun SEMs (FEGSEMs) now approaches the 1 nm level, this value will vary significantly with accelerating voltage, beam current and working distance. The spot will be largest for low voltages, for high beam currents and long working distances, which are unfortunately the conditions most desirable to achieve a high resolution, high intensity CL signal while leaving sufficient clearance for a light collector. The difficulty of achieving a small spot at low voltage can now be helped by the use of aberration correction in the objective lens, which has now been implemented in an SEM [3]. However, for machines with uncorrected lenses and thermionic tungsten filament sources, the spot size may still be the limiting factor in pushing to resolutions below a few 100s of nanometres. 2.1.2 Interaction volume. Generally more significant than the spot size is the lateral spread of the beam during its interaction with the sample bulk. Such interactions are commonly modelled using Monte Carlo electron trajectory simulations (e.g. [4-5]). Figure 1 shows the results of simulating the trajectories of 200 electrons impinging on a 1 nm spot on the surface of a gallium nitride (GaN) sample, repeated for accelerating voltages of 1, 3 and 10 kV. This shows the strongly non-linear dependence of the interaction volume on the incident electron energy for a thick sample. It also illustrates the inherent link between the lateral resolution and probe depth, with these two quantities being approximately equal. This observation is important when pushing the technique towards the nanoscale: it is clearly not possible to achieve nm-scale resolution without limiting the probed volume to the near-surface region, which in turn will generally not be characteristic of the bulk material. Although this may be disadvantageous in some cases, it can also have advantages when deliberately probing near- surface features such as quantum dots and wells [6], or mapping monolayer crystal growth islands (in which the surface sensitivity allows essentially atomic scale depth resolution [7]). If cathodoluminescence is instead carried out on a thin electron-transparent foil in a scanning transmission electron microscope (STEM) geometry [8] (in either an SEM or TEM instrument), then beam spreading within the sample becomes negligible for high accelerating voltages, and this mechanism no longer presents a practical limit to the spatial resolution achievable. Cathodoluminescence nano-characterization of semiconductors Figure 1: Monte Carlo electron trajectory simulations for (left to right) 1 keV, 3 keV and 10 keV beams impinging on a GaN sample. 2.1.3 Carrier diffusion. Since the CL signal results from the recombination of charge carriers, it is clear that the resolution of the technique will not depend directly on the spatial distribution of their generation, but on their steady state distribution after carrier transport has been taken into consideration. Within high-purity semiconductors, it might be thought that the relatively long diffusion lengths compared with the dimensions of the carrier generation function would result in diffusion becoming the dominant factor in limiting the resolution. However, experimentally-confirmed calculations by Donolato [9] suggest that diffusion does not alter the carrier distribution under electron beam excitation sufficiently to significantly affect the image resolution. By numerically simulating the effect of drift from a generation volume approximated by a uniform sphere, it was found that the resolution of the electron beam-induced current (EBIC) technique in imaging localized defects was dependent primarily on the carrier generation profile, even when assuming an infinite diffusion length, L. This is due to the carrier density decreasing in the 3-dimensional case as (1/ r)exp(-r/L) with distance r from a generation source, rather than as the more familiar 1-dimensional form exp(-r/L). This conclusion can be applied equally to CL, which is similar to EBIC in depending on the excess carrier concentration profile for its spatial resolution. While diffusion lengths have been inferred from CL intensity profiles through extended defects [10], from voltage- dependent CL in layered structures [11], and by varying the distance between carrier generation and light collection [12], the effect on image resolution is small when compared with the generation volume. Another consideration for the role of diffusion in CL resolution is that the diffusion length is a property of the bulk material. Most contrast features in CL, however, must by their nature result in a locally reduced diffusion length: for example, if carriers are more likely to recombine radiatively at a given